Public Affairs

National AI strategies are colliding with local power grids, water systems, land-use decisions and public trust. The organizations that earn permission to build will be those that replace generic promises with a verifiable bargain about resources, benefits and long-term responsibility.

By Frank Farnel | Responsible Public Affairs | September 10, 2026

Executive summary

• Data centers are strategic infrastructure, but they are not placeless. Their benefits may be national or global while their electricity demand, water use, noise, land take and grid costs are experienced locally.
• The International Energy Agency estimates that data centers used about 415 terawatt-hours of electricity worldwide in 2024 and could use around 945 TWh in 2030. The global share remains modest; concentration makes the local consequences much larger.^[1]
• France has made AI capacity an industrial and sovereignty priority. Yet RTE projections reported in July 2026 put data-center demand at 23–28 TWh by 2035, compared with about 10 TWh today, while local disputes are increasing around land, electricity and cooling water.^[6]
• Ireland shows what happens when a successful cluster becomes a system-level load: data centers accounted for 23 percent of all metered electricity consumption in 2025, up from 5 percent in 2015. That figure describes metered consumption, not a share of generation, and should be reported as such.^[7]
• Virginia demonstrates both sides of the bargain. Data centers generate construction activity, high-paying operating roles and substantial local tax revenue. A typical 250,000-square-foot facility, however, may employ only about 50 full-time workers after construction, while the state audit forecasts major grid and rate pressures from unconstrained demand.^[8]
• South Africa’s 2026 human-rights inquiry shows the cost of allowing an information vacuum to form. The national Human Rights Commission sought evidence on water, electricity, land, participation and other impacts; by early September, it had received more than 250 submissions, according to the Associated Press.^[9][10]
• The public-affairs task is not to manufacture acceptance. It is to build a credible local resource bargain. Four tests matter: physical capacity, fair distribution, verifiable evidence and durable obligations.

The cloud has acquired a physical address

For years, the language of the digital economy encouraged a category error. Data lived in a cloud. Computing scaled almost frictionlessly. Services arrived through screens, while the infrastructure behind them remained remote, technical and largely invisible.

Artificial intelligence has ended that illusion. A data center is a heavy industrial installation with an unusual product. It requires land, a large and exceptionally reliable electricity connection, cooling, telecommunications, backup systems, construction capacity and a planning decision. The servers may serve users around the world. The substation, water system, access road and neighbors are local.

This distinction is becoming politically decisive. On September 8, 2026, Reuters reported that data centers’ share of electricity use in Peninsular Malaysia reached 9.3 percent during a hot spell in mid-August, compared with an average of about 7 percent for the year. Higher temperatures increased cooling demand. Officials still judged the near-term system manageable, but the episode showed how a national investment success can quickly become a local capacity question when weather, generation and concentrated computing load interact.^[12]

Two days earlier, the debate looked different in South Africa. Civil-society groups were asking for a pause in new hyperscale development until water, land and electricity impacts could be examined. Industry representatives argued that newer facilities use renewable electricity and more efficient cooling. The government sees a significant digital-economy opportunity. All three positions can contain legitimate evidence. The public-affairs failure begins when one side behaves as though the other two have no standing.

National governments naturally emphasize sovereignty, investment and competitiveness. Operators emphasize efficiency, resilience and the services their facilities enable. Local authorities must answer different questions: Which feeder or substation will carry the load? Who pays for expansion? What happens during drought? How many permanent jobs remain after construction? Can the hum be heard at night? What will be disclosed after the permit is granted?

Those are not secondary communications issues. They determine whether a technically feasible project retains political permission, whether an elected official can defend it, and whether the next project enters a receptive or hostile environment.

The scale is global; the constraint is local

The IEA’s 2025 Energy and AI report provides the essential global frame. It estimated data-center electricity consumption at roughly 415 TWh in 2024, about 1.5 percent of worldwide electricity use, and projected around 945 TWh in 2030 in its base case. AI is the principal driver of growth, but the range remains uncertain because efficiency, utilization, investment, grid availability and the pace of adoption can change the outcome.^[1]

Global percentages can reassure and mislead at the same time. A load can be small in the world total and enormous on one distribution network. The 2024 Lawrence Berkeley National Laboratory study estimated that U.S. data centers used 176 TWh in 2023, about 4.4 percent of national electricity, and projected a range of 325–580 TWh in 2028, or 6.7–12 percent. The width of that range is not an analytical defect. It is a warning against presenting one forecast as contracted demand.^[2]

The European Union has begun to convert invisibility into reporting. The recast Energy Efficiency Directive requires information from data centers with installed information-technology power demand of at least 500 kilowatts, and the Commission’s 2024 delegated regulation established a common European reporting database and defined sustainability indicators.^[3][4] In June 2026, the Commission also announced work toward minimum energy-performance standards, with a needs assessment due by 2027; EU data-center capacity was forecast to rise from roughly 12 GW in 2025 to 28 GW in 2030.^[5]

Reporting is necessary, but it is not the same as local legitimacy. Power usage effectiveness can show how much facility energy supports computing rather than cooling and other overhead. It does not reveal whether the grid reinforcement crowds out another project, whether new generation is truly additional, or whether a municipality has exchanged scarce land for a durable economic return. A technically efficient facility can still represent a poor territorial bargain.

That gap defines the role of responsible public affairs. It must connect national strategy, corporate engineering and local consequence before public positions harden. Once trust collapses, another dashboard rarely restores it.

The theory: from social license to a local resource bargain

The phrase social license to operate emerged from extractive industries, where a legal permit could coexist with sustained community opposition. Research by Kieren Moffat and Airong Zhang found that perceived procedural fairness, the quality of contact and confidence in governance influenced community acceptance through trust.^[13] Data centers are not mines, and the analogy should not be pushed too far. They generally create different environmental risks, occupy different supply chains and deliver services with broad social value.

But the governance lesson travels well: formal authorization is not the same as durable legitimacy.

The OECD’s Recommendation on the Governance of Infrastructure makes a related point from the public side. Infrastructure choices should be grounded in evidence, transparent decision making, stakeholder participation and a life-cycle view of cost and performance.^[14] That is a better starting point than a campaign designed to “win the narrative.”

For data centers, I propose a practical Local License Test. It asks four questions.

1. Capacity: can the place carry the project? This covers electricity generation and networks, water availability, wastewater, land, transport, construction labor, emergency services, noise and cumulative impacts. The relevant baseline is the territory, not the operator’s global efficiency average.
2. Distribution: who receives the benefits and who carries the costs? Construction jobs, permanent roles, tax revenue, heat reuse, digital capacity and local procurement belong on one side. Grid reinforcement, public incentives, land opportunity cost, rate exposure, noise and environmental risk belong on the other. National benefit does not cancel local burden.
3. Verification: can an outsider test the promise? A credible commitment has a baseline, unit, boundary, timetable, owner and reporting method. “Powered by renewables” is incomplete without stating whether supply is hourly matched, annual, additional, local or certificate-based. “Water positive” says little about a stressed catchment unless withdrawals and replenishment are geographically and seasonally connected.
4. Durability: does the bargain survive change? Loads ramp up, ownership changes, droughts occur, equipment becomes denser and projects may be canceled after grid assets are built. Permit conditions, tariffs, bonds, monitoring and community-benefit arrangements must address operation, expansion, underuse and exit—not only groundbreaking day.

The four tests are connected. Capacity without distribution may be technically sound and politically brittle. Distribution without verification becomes a brochure. Verification without durable obligations produces a transparent snapshot and an ungoverned future. Durability without capacity simply locks in a bad assumption.

Most importantly, the framework changes the function of engagement. Residents are not an audience segment to be persuaded after site selection. Utilities are not vendors who appear once the announcement is ready. Local officials are not delivery channels for a national industrial policy. Each holds knowledge or authority that determines whether the project can work.

Case study: France’s national ambition meets the territorial test

France has strong reasons to pursue data-center investment. Its electricity system is comparatively low-carbon, it has advanced telecommunications infrastructure, and successive governments have framed computing capacity as an issue of industrial competitiveness and European sovereignty. In February 2025, the government said it had identified 35 potential sites for AI data centers. A France–United Arab Emirates framework agreement contemplated a one-gigawatt campus and investment estimated at $30–50 billion.^[15]

The national message was speed: pre-identified land, available power, simplified procedures and a chance to capture the AI investment cycle. The local question is whether each site offers the same quality of bargain.

RTE estimates reported by Le Monde in July 2026 put French data-center electricity consumption at about 10 TWh today, roughly 2 percent of national use, rising to 15–20 TWh in 2030 and 23–28 TWh in 2035. Those are forecasts, not allocated consumption. They also sit inside a system that must electrify transport, buildings and industry. France may have adequate national generation while facing local connection queues, substation constraints and competing uses for network investment.^[6]

The article also documented increasing opposition around land take, soil sealing, water use and limited permanent employment. RTE’s response to speculative connection requests is instructive: it favors “first ready, first connected” rather than allowing an immature project to reserve capacity indefinitely. That is not merely queue management. It is a public-policy response to uncertainty over which announced loads will materialize.

A weak public-affairs approach would repeat national advantages—nuclear power, sovereignty and investment—at every locality. A stronger one would translate the project into territorial evidence. How much power is required at commissioning, at year three and at full build-out? Which network works are dedicated to the facility, and which strengthen the wider area? What is the water technology under summer conditions? Which positions are construction roles, contracted operations and permanent direct employment? What land use is displaced?

Established fact: France is actively competing for AI and data-center investment, and RTE expects material growth in sector demand. Analysis: France’s national electricity advantage will not, by itself, secure individual projects. Local legitimacy will turn on whether each developer can demonstrate additional capacity and a credible territorial return. Hypothesis: sites selected jointly with network operators and local authorities before a public announcement will face fewer late redesigns than projects introduced as nationally strategic faits accomplis. Comparative project data would be needed to test that proposition.

Case study: Ireland turns clustering success into a national constraint

Ireland offers the clearest statistical picture of concentration. On July 7, 2026, the Central Statistics Office reported that data centers used 7,663 GWh in 2025, a 10 percent increase from 2024. Their share of total metered electricity consumption reached 23 percent, up from 5 percent in 2015. Combined urban and rural residential use represented 28 percent in 2025.^[7]

The wording matters. These are meter data covering customers connected to the mains network. The CSO identified data-center meters through business information, known operators, location and high-consumption screening because no single statistical classification captures every facility. The figures can be revised as meters are added or removed. “Twenty-three percent of metered consumption” is precise. “A quarter of Ireland’s power generation” would not be the same claim.

Ireland’s cluster delivered real advantages: connectivity, investment, a mature technology ecosystem and tax revenue. It also accumulated electricity demand in and around Dublin faster than grid capacity could comfortably absorb. Connection restrictions and later rules for large energy users were consequences of success arriving before a durable allocation framework.

The public-affairs lesson is not that Ireland should have rejected data centers. It is that cumulative impact cannot be managed project by project if every applicant describes its own load as marginal. The ninth facility enters a system shaped by the first eight. A commitment to renewable supply must be tested against additional generation, timing, congestion and the need for firm capacity when wind or solar output is low.

Ireland also warns against treating household comparisons as a communications shortcut. Saying data centers use “almost as much as every home” attracts attention, but it does not answer whether the load is flexible, how it affects marginal generation, who finances new infrastructure or what economic value it supports. The correct response is not to avoid the comparison. It is to add the missing system information.

Success: Ireland built a globally significant digital infrastructure cluster. Failure: grid and cumulative-impact governance lagged behind that growth, forcing more restrictive intervention later. Practical lesson: a jurisdiction should define capacity, location, flexibility and cost-allocation rules before connection scarcity turns every new proposal into a zero-sum political event.

Case study: Virginia reveals the uneven economics of the bargain

Northern Virginia is the world’s largest data-center market. The state’s Joint Legislative Audit and Review Commission reported in 2024 that the region represented 13 percent of reported global operating capacity and 25 percent of capacity in the Americas. Fiber, reliable electricity, customers, land and a state tax incentive helped create that position.^[8]

The audit is valuable because it resists a simple pro- or anti-development story. It estimated that the sector contributes 74,000 jobs, $5.5 billion in labor income and $9.1 billion in state GDP annually. Much of that activity, however, comes from construction. A typical 250,000-square-foot data center may have approximately 50 full-time workers, about half contractors, after a construction phase that can employ roughly 1,500 people at its peak for 12–18 months.

Local fiscal benefits vary dramatically. Among five localities with mature markets, data-center revenue ranged from less than 1 percent to 31 percent of total local revenue. That range should end the habit of discussing “the” community benefit. The value depends on tax rates, project scale, existing services and what the locality gives up to secure investment.

The physical findings are equally mixed. The report judged current statewide water use sustainable, while noting that some localities face tighter water conditions and that allocation among uses receives less oversight. It found backup-generator emissions to be a relatively small share of regional pollution under current operating patterns. At the same time, one-third of data centers were near residential areas; inadequate zoning had produced adverse impacts in some cases, and low-frequency noise created a distinctive enforcement problem even when ordinary noise limits were not breached.

The largest risk concerned power. JLARC’s independent forecast found that unconstrained Virginia demand could double within ten years, driven mainly by data centers, and concluded that meeting even half that demand would be difficult. Under existing assumptions, a typical Dominion Energy residential customer could see generation- and transmission-related costs rise by an estimated $14–$37 per month in constant dollars by 2040. This is a modeled range, not a current bill increase or a guaranteed outcome.

Virginia therefore demonstrates what mature public affairs should look like: neither industry boosterism nor impact absolutism. The case for a project strengthens when permanent employment is stated separately from construction, fiscal benefits are shown net of incentives and service costs, water claims are local, noise modeling precedes zoning, and electricity-rate risks are allocated explicitly.

Established fact: Virginia receives substantial economic and fiscal benefits, alongside concentrated land-use and power-system pressures. Analysis: the legitimacy risk comes less from any single impact than from asymmetry—local communities can see the buildings and hear the equipment, while grid financing, tax treatment and contracted load remain difficult to inspect.

Case study: South Africa shows what fills an information vacuum

South Africa has positioned itself as the leading data-center market on the continent. President Cyril Ramaphosa has said the country hosts about 70 percent of Africa’s capacity. That creates a genuine opportunity to anchor cloud services, investment and digital capability.

It also places new industrial demand inside a political economy shaped by electricity interruptions, water insecurity and severe inequality. In May 2026, the South African Human Rights Commission opened a call for submissions on data centers and digital infrastructure. Its stated themes included electricity and water use, climate and environmental justice, community consultation, privacy, cybersecurity and socioeconomic rights.^[9]

Five civil-society organizations submitted a joint report in August and asked the Commission to consider a national inquiry, mandatory disclosure, community-benefit obligations and a temporary pause on new hyperscale approvals. The submission claimed more than 60 known facilities with roughly 500 MW of disclosed capacity and argued that no regulator currently sees the sector’s full footprint. Those figures and conclusions come from an advocacy submission; they are not findings adopted by the Commission.^[11]

By September 4, the Associated Press reported that the Commission had received more than 250 submissions. Civil-society groups pressed for a halt, while industry representatives disputed the implication that data centers cause resource scarcity and pointed to renewable procurement and lower-water cooling technologies. The inquiry had not produced final findings at the research cutoff.^[10]

This is precisely where evidence discipline matters. A project team should not dismiss campaign estimates because they come from opponents. Nor should public authorities repeat them as established fact before review. The correct response is a shared factual baseline: facility register, planned and actual load, cooling method, seasonal withdrawal, land footprint, emergency-generation assumptions, employment and benefit distribution.

The absence of official aggregate data does not create political neutrality. It transfers agenda-setting power to whichever actor publishes the first plausible total. Once that total becomes the reference point, the operator is forced into rebuttal and the regulator appears late.

Established fact: the Human Rights Commission invited evidence and received substantial participation; civil society requested a pause and made quantified claims; industry challenged the wider narrative. Unresolved: the Commission’s conclusions, the verified national aggregate footprint and the appropriate regulatory response. Public-affairs lesson: disclosure delayed until controversy is not transparency. It is crisis response.

Comparative local-license matrix

Jurisdiction	Strategic objective	Evidence of local pressure	Governance response	Public-affairs lesson
France	AI competitiveness, investment and digital sovereignty	Projected demand growth; local disputes over land, water, power and employment	Pre-identified sites and readiness-based grid connection	Translate national advantage into a project-specific territorial bargain before announcement
Ireland	Maintain a major European digital cluster	Data centers reached 23% of metered electricity consumption in 2025	Connection constraints followed by stronger large-user conditions	Govern cumulative load and additionality, not only individual project efficiency
Virginia	Protect a globally dominant market and its tax base	Grid-cost exposure, residential proximity, noise and uneven permanent employment	State audit; proposals on rates, zoning, water estimates and sound modeling	Disaggregate benefits and costs by locality, phase and customer class
South Africa	Lead Africa’s digital infrastructure market	Rights-based concerns over scarce resources, land and transparency	National human-rights inquiry and contested calls for a pause	Create an official baseline before advocacy estimates define the debate

What responsible engagement looks like in practice

The data-center debate attracts two equally unhelpful simplifications. One says communities oppose what they do not understand. The other says every large facility is an extractive bargain imposed by a global technology company. Both erase the evidence that matters.

Some opposition will remain even after a rigorous process. Residents may reasonably prefer housing, agriculture or lower-intensity industry on a site. An elected authority may decide that national digital capacity outweighs those objections. Responsible public affairs does not promise consensus. It makes the trade-off legible, ensures affected people can test the evidence and leaves a record of why the decision was taken.

That requires engagement to start with engineering and finance, not messaging. Public-affairs teams need access to load curves, connection studies, cooling design, tax terms, employment assumptions and construction schedules. If commercially sensitive information cannot be published, the organization should explain the category, legal basis and independent assurance used—not label the entire model confidential.

It also requires cumulative analysis. A company may accurately say its facility uses less water than an older design. A community may accurately say the regional total is rising. Both claims can be true. The relevant decision concerns the facility inside the catchment, grid and development pipeline—not the facility compared with a technological past.

Finally, engagement must survive commissioning. The people who negotiated the permit may leave. The site may change tenants. Actual computing load can differ from the announced nameplate capacity. A durable mechanism publishes performance, handles complaints, triggers corrective action and reviews community benefits over the operating life.

What Leaders Should Do Now

1. Put local legitimacy on the investment committee agenda. Require a Local License Test before land acquisition becomes irreversible. Review capacity, distribution, verification and durability alongside return, power price and time to connection.
2. Publish three demand cases. Show expected electricity and water use at commissioning, normal operation and full build-out. Distinguish connected capacity from actual consumption and annual totals from peak demand.
3. State the counterfactual. Explain what grid, water or road investment would occur without the project; who funds the incremental work; and who bears stranded costs if the load arrives late or never reaches forecast scale.
4. Separate jobs by phase. Report construction job-years, peak contractors, permanent direct roles, outsourced roles and induced employment separately. Do not let a temporary peak become the implied operating workforce.
5. Make environmental claims territorial. Report withdrawals, consumption, discharge, energy source and emissions at the relevant facility and system boundary. Match water replenishment to the catchment and renewable claims to time and additionality.
6. Design with neighbors before seeking applause. Complete sound modeling, visual-impact analysis, traffic planning and emergency coordination early enough for local knowledge to change the design.
7. Create a public obligations register. List each permit condition and voluntary promise, its owner, metric, reporting interval, independent verifier and remedy. Update it after ownership or design changes.
8. Give public affairs the right to stop an announcement. If the engineering baseline, benefit package or local authority coordination is incomplete, delay the launch. Surprise converts correctable project questions into questions of respect.
9. Prepare for legitimate disagreement. Engagement is not a promise that every stakeholder will agree. Record unresolved positions faithfully, explain the decision criteria and preserve a route for monitoring, appeal and correction.

Conclusion: infrastructure earns trust in the place where it lands

The digital economy once benefited from distance between service and infrastructure. That distance is closing. AI has made computing capacity strategic, electricity-intensive and visible. The cloud now arrives as a planning application, a substation, a cooling system and a long-term claim on a territory.

France shows the opportunity and the danger of assuming national ambition will carry local consent. Ireland shows how a successful cluster can outgrow the governance designed for it. Virginia shows that benefits are real but uneven, and that construction, employment, tax revenue, noise and grid costs must be separated rather than blended into one economic-impact number. South Africa shows how quickly an information gap becomes a legitimacy gap.

The answer is not a more elegant explanation of why AI matters. Communities already understand that governments and companies consider it important. They want to know what one facility will take, what it will add, who is accountable and whether the bargain will still hold when conditions change.

That is the new public-affairs test. The winning project will not be the one with the loudest national endorsement. It will be the one whose local obligations remain credible after the cameras leave.

Key Evidence

• 415 TWh to 945 TWh: the IEA’s estimate of global data-center electricity use in 2024 and base-case projection for 2030. The 2030 number is a model, not a commitment.^[1]
• 23 percent: data centers’ share of Ireland’s total metered electricity consumption in 2025, up from 5 percent in 2015; actual consumption was 7,663 GWh.^[7]
• 23–28 TWh: RTE’s reported projection for French data-center electricity consumption in 2035, compared with about 10 TWh currently.^[6]
• About 50 operating jobs: the estimated full-time workforce at a typical 250,000-square-foot Virginia data center, compared with roughly 1,500 workers at peak construction.^[8]
• $14–$37 per month: JLARC’s modeled increase in generation- and transmission-related costs for a typical Dominion residential customer by 2040, in constant dollars, under the analyzed growth assumptions.^[8]
• More than 250 submissions: the participation reported by the Associated Press in the South African Human Rights Commission’s 2026 data-center inquiry; no final findings had been issued at the research cutoff.^[10]

Glossary

Connected capacityThe maximum electrical load a facility is authorized or equipped to draw. It should not be confused with actual consumption.Hyperscale data centerA very large computing facility designed to expand rapidly and support high-volume cloud, platform or AI workloads.LoadThe amount of electrical power demanded at a particular moment, usually expressed in megawatts. Energy consumed over time is expressed in megawatt-hours or terawatt-hours.AdditionalityThe degree to which a project causes new generation, network capacity, water replenishment or other resources to exist beyond what would have occurred anyway.Power usage effectiveness (PUE)A ratio comparing all energy used by a data center with the energy used by its computing equipment. Lower values generally indicate less overhead, but PUE does not measure total load or local system impact.Water usage effectiveness (WUE)A facility metric relating water use to computing energy. Results depend on boundaries, cooling technology, climate and whether direct and indirect water are included.Stranded costInfrastructure expenditure that must still be recovered after expected demand fails to materialize or an asset becomes underused.Social license to operateAn informal level of acceptance and trust held by affected communities and stakeholders, distinct from a legal permit.

References and Further Reading

Official and institutional sources

1. International Energy Agency, “Energy and AI,” IEA, April 10, 2025.
2. Arman Shehabi, Sarah J. Smith, Andrew Hubbard, Alexandra Newkirk and Nuoa Lei, “2024 United States Data Center Energy Usage Report,” Lawrence Berkeley National Laboratory, December 2024.
3. European Parliament and Council, “Directive (EU) 2023/1791 on Energy Efficiency,” Official Journal of the European Union, September 13, 2023.
4. European Commission, “Commission Delegated Regulation (EU) 2024/1364 on the First Phase of a Common Union Rating Scheme for Data Centres,” Official Journal of the European Union, March 14, 2024.
5. Central Statistics Office of Ireland, “Data Centres Metered Electricity Consumption 2025: Key Findings,” July 7, 2026.
6. Virginia Joint Legislative Audit and Review Commission, “Data Centers in Virginia,” Commonwealth of Virginia, 2024.
7. South African Human Rights Commission, “Call for Public Submissions on the Human Rights Impact of Data Centres and Digital Infrastructure,” May 22, 2026.
8. Organisation for Economic Co-operation and Development, “Recommendation of the Council on the Governance of Infrastructure,” OECD/LEGAL/0460, July 17, 2020.

Academic and theoretical work

9. Kieren Moffat and Airong Zhang, “The Paths to Social Licence to Operate: An Integrative Model Explaining Community Acceptance of Mining,” Resources Policy, Vol. 39, 2014.

Authoritative reporting and case materials

10. Reuters, “EU Plans Energy Standards for Data Centres Amid Concerns Over Soaring Power Use,” June 3, 2026.
11. Le Monde, “France Strives to Keep Up in Global Data Center Race as Opposition Mounts,” July 27, 2026.
12. Associated Press, “Civil Rights Groups Urge a Halt to South Africa Data Centers Boom Amid Water and Power Fears,” September 4, 2026.
13. Open Secrets, Housing Assembly, Foxglove, Research + Action and Planetary AI Collective, “Submission to the South African Human Rights Commission: Data Centres and Human Rights in South Africa,” press release and submission summary, August 25, 2026.
14. Reuters, “Malaysia’s Data Centres Guzzle More Power as Temperatures Soar, Officials Say,” September 8, 2026.
15. Reuters, “France, UAE Agree to Develop 1 Gigawatt AI Data Centre,” February 6, 2025.

Source and methodology note

Research was completed at 6:00 a.m. Central European Summer Time on September 10, 2026. The article prioritizes official legislation, statistical releases, public audits, institutional inquiries and major wire-service reporting. Global and national electricity projections are scenarios rather than observed demand and are labeled accordingly. Electricity figures use different boundaries: Ireland’s 23 percent is a share of total metered consumption identified by the CSO; France’s 2030 and 2035 figures are forecasts attributed to RTE; the IEA figure is a modeled global total. They should not be compared without those distinctions.

The South African Human Rights Commission had not issued final findings at the cutoff. Claims from the civil-society submission are therefore identified as advocacy claims, not official conclusions. The article does not attempt a facility-by-facility environmental assessment. Water use, carbon intensity, employment and fiscal effects vary significantly by site, climate, technology, grid and tax regime. “Established fact,” “analysis,” “hypothesis” and “unresolved” labels are used where the evidentiary status could otherwise be confused.

Suggested internal links

• If You Cannot Trace the Answer, It Was Not Engagement: The New Public Affairs Standard — link from the discussion of public obligations registers and transparent disposition of community concerns.
• Trust: The Key Asset in Public Affairs — link from the conceptual discussion of durable legitimacy.
• Closing the Gap: Strategic Communications in Public Affairs — link from the section on translating engineering evidence into public decisions.
• Business Diplomacy vs. Government Relations: Key Differences — link from the discussion of national strategy and territorial stakeholders.